基于生成对抗网络的负荷序列随机场景生成方法

A Stochastic Scenario Generation Method of Load Series Based on Generative Adversarial Network

随着可再生能源逐步渗透,电力系统随机性不断加强,其不确定性为调度、规划、运行带来了更大的挑战,因此需要研究针对不确定进行建模的方法。提出一种基于生成对抗网络的负荷场景随机生成方法,该方法基于深度卷积生成对抗网络架构,以JS散度作为目标函数,对生成器以及判别器交替进行训练。针对生成负荷序列质量的衡量,从数据多样性以及锐度2个方面,提出TSTR(train on synthetic test on real)以及TRTS(test on real train on synthetic)2个指标,基于支撑向量回归模型进行判断,实验结果表明,随着训练的进行,生成器产生的数据质量逐渐提高,且当训练完成时可以产生满足多样性以及锐度要求的数据。

With the gradual penetration of renewable energy,the randomness of power systems continues to increase,and its uncertainty poses greater challenges for scheduling,planning,and operation.Therefore,it is necessary to study methods for modeling uncertainties.A random generation method of load series based on generative adversarial network is proposed.This method is based on deep convolution generative adversarial network architecture.JS divergence is used as the objective function,and the generator and discriminator are alternately trained.To evaluate the quality of generated data,two indexes of TSTR and TRTS are proposed from the aspects of data diversity and sharpness.The results are judged based on the support vector regression model.The experimental results show that with the pass of the training process,data quality is gradually improved,and when the train is completed,data that satisfies diversity and sharpness requirements can be generated.